Automatically determining a current value for a home

Abstract

A facility for valuing a distinguished home located in a distinguished geographic area is described. The facility receives home attributes for the distinguished home. For each of a plurality of valuation sub-models, the facility applies the valuation sub-model to the received home attributes to obtain a sub-model valuation for the distinguished home. The facility further applies a meta-model to the record home attributes to obtain a relative weighting factor for each sub-model. The facility then uses the obtained relative weighting factors to combine the sub-model valuations to obtain an overall valuation for the distinguished home. The facility reports the obtained valuation for the distinguished home.

We claim: 
     
       1. A method, performed by a computing system having a memory and a processor, for valuing a distinguished home among a population of homes each having attributes, the method comprising:
 training, by the computing system, a plurality of valuation sub-models, each valuation sub-model being capable of producing a valuation for a home among the population of homes based upon attributes of the home, at least in part by, for each of one or more of the plurality of valuation sub-models,
 for each of a number of homes recently sold in a geographic region, identifying, by the computing system, attributes of the home and its selling price, and 
 constructing, by the computing system, based upon one or more of the identified attributes and one or more of the identified selling prices, a forest of classification trees in which each non-leaf node represents a basis for differentiating homes based upon one of the identified attributes, 
 each valuation sub-model being based on information about recent sales of homes among the population and attributes of the recently sold homes; 
 
 after training the plurality of valuation sub-models,
 for each of two or more trained valuation sub-models,
 producing, by the computing system, a valuation for the distinguished home at least in part by applying the trained valuation sub-model to attributes of the distinguished home; 
 
 
 retrieving, by the computing system, a valuation meta-model based on information about recent sales of homes among the population and attributes of the sold homes, wherein the valuation meta-model specifies a relative weighting factor for each trained valuation sub-model of the plurality of trained valuation sub-models and wherein each relative weighting factor is based at least in part on a median error value for a corresponding trained valuation sub-model; 
 after retrieving the valuation meta-model,
 obtaining a relative weighting factor for each of the two or more trained valuation sub-models at least in part by applying the retrieved valuation meta-model to attributes of the distinguished home, so that a first relative weighting factor is obtained for a first trained valuation sub-model of the two or more trained valuation sub-models and a second relative weighting factor is obtained for a second trained valuation sub-model of the two or more trained valuation sub-models; 
 
 after obtaining the relative weighting factors for each of the two or more trained valuation sub-models and producing the valuations for each of the two or more trained valuation sub-models,
 combining, by the computing system, the produced valuations of the distinguished home in accordance with the relative weighting factors obtained for the two or more trained valuation sub-models to obtain an overall valuation for the distinguished home; 
 
 after obtaining the overall valuation for the distinguished home,
 performing bias adjustments by scoring at least one attribute of the distinguished home with a systematic error model to obtain an expected percentage deviation of the overall valuation from an actual value of the distinguished home, 
 adjusting the obtained overall valuation to correct for the expected percentage deviation, and 
 smoothing on the adjusted obtained overall valuation by replacing the adjusted obtained overall valuation with a weighted average of the adjusted obtained overall valuation and an overall valuation from a previous valuation cycle; 
 
 receiving, from a user, one or more email addresses; and 
 in response to receiving the one or more email addresses from the user, automatically triggering the sending of a message containing a uniform resource locator used to access a valuation for the distinguished home to the one or more email addresses. 
 
     
     
       2. The method of  claim 1 , further comprising, prior to applying the trained valuation sub-models:
 identifying an attribute of the distinguished home for which a value is not available; and 
 applying a model that predicts the value of the identified attribute based upon the values of other attributes for the population of homes to obtain an imputed value for the identified attribute. 
 
     
     
       3. The method of  claim 1 , wherein each leaf node of a first tree from among the forest of classification trees includes a list of selling prices for a plurality of homes represented by the leaf node. 
     
     
       4. The method of  claim 3 , wherein obtaining a valuation for the distinguished home from the first tree from among the forest of classification trees comprises determining a median value from among the list of selling prices for the plurality of homes represented by a leaf node corresponding to the distinguished home. 
     
     
       5. The method of  claim 1 , wherein constructing the forest of classification trees further comprises, for each tree,
 for at least one node of the tree,
 determining that the at least one node should be split at least in part by determining that the number of sales represented by the at least one node satisfies a split threshold; 
 in response to determining that the at least one node should be split,
 creating a pair of child nodes for the at least one node, each child node representing one or more sales of recently sold homes and an attribute subrange on a different side of a selected split point. 
 
 
 
     
     
       6. The method of  claim 1 , wherein constructing the forest of classification trees further comprises, for each tree,
 for at least one node of the tree,
 determining that the at least one node should not be split at least in part by determining that the number of sales represented by the at least one node does not satisfy a split threshold; 
 in response to determining that the node should not be split,
 determining a mean selling price of sales represented by the node. 
 
 
 
     
     
       7. The method of  claim 1 , further comprising:
 submitting, to a linear regression engine, attributes for a plurality of homes recently sold in the geographic region; 
 receiving, from the linear regression engine, a set of coefficients representing a linear valuation formula; and 
 applying the linear valuation formula to attributes of the distinguished home to obtain a valuation for the distinguished home. 
 
     
     
       8. The method of  claim 1 , wherein a first tree of the forest of classification trees differentiates homes according to a number of bedrooms in each home and wherein a second tree of the forest of classification trees differentiates homes according to a number of bathrooms in each home. 
     
     
       9. The method of  claim 8 , wherein a third tree of the forest of classification trees differentiates homes according to an improved floor area of each home and wherein a fourth tree of the forest of classification trees differentiates homes according to a lot size of each home. 
     
     
       10. A computing system, having a memory and a processor, for valuing a distinguished home among a population of homes each having attributes, the computing system comprising:
 a component configured to train a plurality of valuation sub-models, each valuation sub-model being based on information about recent sales of homes among the population and attributes of the sold homes, each valuation sub-model being capable of producing a valuation for a home among the population of homes based upon attributes of the home, comprising: 
 a component configured to, for each valuation sub-model,
 for each of a number of homes recently sold in a geographic region, identify attributes of the home and its selling price; 
 
 a component configured to, after the plurality of valuation sub-models are trained,
 for each of two or more trained valuation sub-models,
 produce a valuation for the distinguished home at least in part by applying the trained valuation sub-model to attributes of the distinguished home; 
 
 
 a component configured to retrieve a valuation meta-model based on information about recent sales of homes among the population and attributes of the sold homes, wherein the valuation meta-model specifies a relative weighting factor for each trained valuation sub-model of the plurality of trained valuation sub-models and wherein each relative weighting factor is based at least in part on a median error value for a corresponding trained valuation sub-model: 
 a component configured to, after the valuation meta-model is retrieved, obtain a relative weighting factor for each of the two or more trained valuation sub-models at least in part by applying the retrieved valuation meta-model to attributes of the distinguished home, so that a first relative weighting factor is obtained for a first trained valuation sub-model of the two or more trained valuation sub-models and a second relative weighting factor is obtained for a second trained valuation sub-model of the two or more trained valuation sub-models; 
 a component configured to, after the relative weighting factors for each of the two or more trained valuation sub-models are obtained and the valuations for the distinguished home are produced, combine the produced valuations for the distinguished home in accordance with the relative weighting factors obtained for the two or more trained valuation sub-models to obtain an overall valuation for the distinguished home; and 
 a component configured to, after the overall valuation for the distinguished home is obtained,
 adjust the obtained overall valuation to correct for expected percentage deviation, and 
 
 smooth on the adjusted obtained overall valuation by replacing the adjusted obtained overall valuation with a weighted average of the adjusted obtained overall valuation and an overall valuation from a previous valuation cycle; 
 in response to receiving the one or more email addresses from the user automatically triggering the sending of a message containing a uniform resource locator used to access a valuation for the distinguished home to the one or more email addresses; 
 wherein each of the components comprises computer-executable instructions stored in the memory for execution by the processor. 
 
     
     
       11. The computing system of  claim 10 , wherein the homes used in the training of at least one valuation sub-model are randomly selected. 
     
     
       12. The computing system of  claim 10 , further comprising:
 a component configured to identify an attribute of the distinguished home for which a value is not available; and 
 a component configured to impute a value for the value that is not available. 
 
     
     
       13. The computing system of  claim 12 , wherein an imputed value for a first attribute is based upon a median value for that attribute from among the homes recently sold in the geographic region. 
     
     
       14. The computing system of  claim 12 , wherein an imputed value for a first attribute is based upon a mode value for the first attribute from among the homes recently sold in the geographic region, wherein the mode value for the first attribute from among the homes recently sold in the geographic region corresponds to the most frequent value for the first attribute from among the homes recently sold in the geographic region. 
     
     
       15. The computing system of  claim 10 , wherein at least one of the valuation sub-models comprises a forest of classification trees and a first tree of the forest of classification trees differentiates homes according to a number of floors in each home and wherein a second tree of the forest of classification trees differentiates homes according to a year in which each home was constructed. 
     
     
       16. The computing system of  claim 10 , wherein at least one of the valuation sub-models comprises a forest of classification trees and a first tree of the forest of classification trees differentiates homes according to a roof type of each home and wherein a second tree of the forest of classification trees differentiates homes according to an indication of a primary use of each home. 
     
     
       17. The method of  claim 1 , further comprising:
 for each of the one or more of the plurality of valuation sub-models,
 for each tree of the forest of classification trees constructed for the valuation sub-model,
 rating a usefulness of the tree by applying the tree to homes other than homes that were selected to construct the tree, and 
 weighting the tree based at least in part on the rating. 
 
 
 
     
     
       18. The method of  claim 1 , wherein applying, by the computing system, the valuation meta-model to attributes of the distinguished home to obtain a relative weighting factor for each of the two or more trained valuation sub-models comprises:
 obtaining the first relative weighting factor for valuations produced by the first trained valuation sub-model of the two or more trained valuation sub-models; 
 obtaining the second relative weighting factor for valuations produced by the second trained valuation sub-model of the two or more trained valuation sub-models, wherein the second relative weighting factor is different from the first relative weighting factor; and 
 obtaining a third relative weighting factor for valuations produced by a third trained valuation sub-model of the two or more trained valuation sub-models, wherein the third relative weighting factor is different from the first relative weighting factor and wherein the third relative weighting factor is different from the second relative weighting factor. 
 
     
     
       19. A method, performed by a computing system having a memory and a processor, for valuing a distinguished home among a population of homes each having attributes, the method comprising:
 training, by the computing system, a plurality of valuation sub-models, each valuation sub-model being based on information about recent sales of homes among the population and attributes of the sold homes, each valuation sub-model being capable of producing a valuation for a home among the population of homes based upon attributes of the home; 
 retrieving, by the computing system, a valuation meta-model based on information about recent sales of homes among the population and attributes of the sold homes, wherein the valuation meta-model specifies a relative weighting factor for each trained valuation sub-model of the plurality of trained valuation sub-models and wherein each relative weighting factor is based at least in part on a median error value for a corresponding trained valuation sub-model; 
 after obtaining relative weighting factors for each of two or more trained valuation sub-models at least in part by applying the retrieved valuation meta-model to attributes of a distinguished home and producing valuations for the distinguished home for each of the two or more trained valuation sub-models,
 combining, by the computing system, the produced valuations of the distinguished home in accordance with the relative weighting factors obtained for the two or more trained valuation sub-models to obtain an overall valuation for the distinguished home; 
 
 after obtaining the overall valuation for the distinguished home,
 smoothing on the obtained overall valuation by replacing the obtained overall valuation with a weighted average of the obtained overall valuation and an overall valuation from a previous valuation cycle; 
 
 receiving, from a user, one or more email addresses; and 
 in response to receiving the one or more email addresses from the user, automatically triggering the sending of a message containing a uniform resource locator used to access a valuation for the distinguished home to the one or more email addresses. 
 
     
     
       20. The method of  claim 19 , further comprising:
 scoring, by the computing system, each of the three or more of the plurality of trained valuation sub-models at least in part by, 
 for each of three or more of the plurality of trained valuation sub-models, applying, by the computing system, the trained valuation sub-models to each of the sold homes, such that each trained valuation sub-model produces a valuation for each of the sold homes. 
 
     
     
       21. The method of  claim 19 , further comprising:
 for each of three or more of the plurality of trained valuation sub-models,
 applying, by the computing system, the trained valuation sub-model to attributes of the distinguished home to obtain a valuation for the distinguished home; 
 
 applying, by the computing system, the valuation meta-model to attributes of the distinguished home to obtain at least a first relative weighting factor for a first trained valuation sub-model, a second relative weighting factor for a second trained valuation sub-model, and a third relative weighting factor for a third trained valuation sub-model; and 
 combining, by the computing system, the valuations of the distinguished home obtained from the valuation sub-models in accordance with the obtained relative weighting factors to obtain an overall valuation for the distinguished home. 
 
     
     
       22. The method of  claim 19 , further comprising:
 receiving, from a user, an indication of at least one home comparable to the distinguished home; 
 making a copy of a recent sales table for a geographic region that includes the distinguished home and was used to construct at least one valuation sub-model; and 
 altering the copy of the recent sales table to increase a weighting in the copy of the recent sales table of the at least one home comparable to the distinguished home. 
 
     
     
       23. The method of  claim 19 , wherein altering the copy of the recent sales table to increase the weighting in the copy of the recent sales table of the at least one home comparable to the distinguished home comprises adding copies of rows corresponding to the at least one home comparable to the distinguished home to the copy of the recent sales table. 
     
     
       24. A computer-readable medium storing instruction that, when executed by a computing system having a memory and a processor, cause the computing system to perform operations for valuing a distinguished home among a population of homes each having attributes, the operations comprising:
 training, by the computing system, a plurality of valuation sub-models, each valuation sub-model being based on information about recent sales of homes among the population and attributes of the sold homes, each valuation sub-model being capable of producing a valuation for a home among the population of homes based upon attributes of the home; 
 retrieving, by the computing system, a valuation meta-model based on information about recent sales of homes among the population and attributes of the sold homes, wherein the valuation meta-model specifies a relative weighting factor for each trained valuation sub-model of the plurality of trained valuation sub-models and wherein each relative weighting factor is based at least in part on a median error value for a corresponding trained valuation sub-model; 
 after obtaining relative weighting factors for each of two or more trained valuation sub-models at least in part by applying the retrieved valuation meta-model to attributes of a distinguished home and producing valuations for the distinguished home for each of the two or more trained valuation sub-models,
 combining, by the computing system, the produced valuations of the distinguished home in accordance with the relative weighting factors obtained for the two or more trained valuation sub-models to obtain an overall valuation for the distinguished home; 
 
 smoothing on the obtained overall valuation by replacing the obtained overall valuation with a weighted average of the obtained overall valuation and an overall valuation from a previous valuation cycle; 
 receiving, from a user, one or more email addresses; and 
 in response to receiving the one or more email addresses from the user, automatically triggering the sending of a message containing a uniform resource locator used to access a valuation for the distinguished home to the one or more email addresses. 
 
     
     
       25. A method in a computing system for valuing a distinguished home among a population of homes each having attributes, the method comprising:
 retrieving a plurality of valuation sub-models, each valuation sub-model being based on information about recent sales of homes among the population and attributes of the sold homes, each valuation sub-model being capable of producing a valuation for a home among the population of homes based upon attributes of the home; 
 retrieving a valuation meta-model based on information about recent sales of homes among the population and attributes of the sold homes, the valuation meta-model being capable of specifying, for a combination of the valuations produced for a home among the plurality of homes by each of the plurality of sub-models, the relative weight to be given to the valuation produced for the home by each of the plurality of sub-models; 
 for each of the plurality of valuation sub-models, applying the valuation sub-model to attributes of the distinguished home to obtain a valuation of the distinguished home; 
 applying the valuation meta-model to attributes of the distinguished home to obtain a weighting factor for each of the valuation sub-models; 
 combining the valuations of the distinguished home obtained from the valuation sub-models in accordance with the obtained weighting factors to obtain an overall valuation for the distinguished home; and 
 smoothing on the obtained overall valuation by replacing the obtained overall valuation with a weighted average of the obtained overall valuation and an overall valuation from a previous valuation cycle; 
 in response to receiving the one or more email addresses from the user automatically triggering the sending of a message containing a uniform resource locator used to access a valuation for the distinguished home to the one or more email addresses.